Related Experiment Video
Updated: May 28, 2025
07:43
In Vivo Quantification of Hip Arthrokinematics during Dynamic Weight-bearing Activities using Dual Fluoroscopy
Published on: July 2, 2021
3.0K
Deep learning based screening model for hip diseases on plain radiographs
Jung-Wee Park1,2, Seung Min Ryu3, Hong-Seok Kim1,4
1Department of Orthopaedic Surgery, Seoul National University College of Medicine, Seoul, South Korea.
Plos One
|February 13, 2025
Summary
A new deep learning model accurately screens hip diseases from plain radiographs. This AI tool shows high reliability, potentially improving diagnostic accuracy for physicians.
Area of Science:
- Radiology
- Artificial Intelligence
- Medical Imaging
Background:
- Interpretation of hip radiographs varies among physicians.
- Developing reliable screening tools for hip diseases is crucial.
Purpose of the Study:
- To develop and validate a deep learning model for distinguishing normal hips from severe hip diseases using plain radiographs.
Main Methods:
- Utilized 1,726 hip radiographs (500 normal, 1,226 hip disease) from 2004-2012.
- Trained and tested four deep learning models, including variations in image preprocessing and backbone algorithms (DenseNet to EfficientNet).
- Evaluated models based on accuracy, sensitivity, specificity, PPV, NPV, F1-score, and AUROC.
Main Results:
- The final model achieved high performance: accuracy 0.96, sensitivity 0.96, specificity 0.97, and AUROC 0.99 on the internal test set.
- External validation demonstrated strong results: accuracy 0.94, sensitivity 0.93, specificity 0.96, and AUROC 0.98.
- GradCAM analysis showed improved focus on relevant anatomical structures in refined models.
Conclusions:
- Deep learning models demonstrate high accuracy and reliability in screening hip diseases from plain radiographs.
- This AI-driven approach can assist physicians in more accurate hip condition diagnosis.
